Language Model

Large language models are essentially stateless functions whose weights are fixed at inference time and do not learn over time.

The Agent Harness does not replace the reasoning capability of the core model, but instead builds a runtime management shell around the model, responsible for cross-cutting concerns such as tool invocation, state management, and error recovery.

Test cases provide AI with a machine-verifiable objective function, converting vague natural language requirements into precise pass/fail signals.

Current mainstream large language models exhibit significant deficiencies in both confidence calibration and selective abstention, tend to be overconfident, and when faced with questions outside the training distribution, tend to produce fluent but incorrect answers.

Large language models cannot structurally distinguish between user instructions and instructions embedded in content, because both are part of the same token sequence from the model's perspective.

The confidence output of large language models is fundamentally a product of probability distributions rather than a genuine judgment of factual truth or falsehood, and may express high confidence in incorrect statements.

Reward model training can use learning-to-rank frameworks such as the Bradley-Terry model, training a neural network to predict the probability of human preference between any two responses, taking a (prompt, response) pair as input and outputting a scalar score.

The context window refers to the length of text a model can remember and reference in a single conversation; information beyond this range is truncated and discarded.

The tool-calling mechanism allows large language models to dynamically invoke external functions during inference, representing a fundamental leap from language models to action models.

A language model itself can only output text and cannot directly perform operations such as opening files or pressing buttons.

This article is compiled from automatically generated content; the original text was drafted by a language model and checked automatically, with no human author credited.

The larger the parameter count, the stronger a model generally is in language understanding and generation quality, but the memory and compute requirements also increase proportionally.

Language models excel at pattern generation, but antibody developability involves multi-dimensional constraints such as affinity, stability and immunogenicity, and purely generated results must undergo rigorous computational screening and wet-lab validation.

Simulating a CPU with a language model faces the dual constraints of context length and state precision; the longer the program, the more the key to the method's practicality is whether the model's attention mechanism can carry discrete states losslessly.

The second stage of the two-stage architecture uses a language model to perform query-specific evidence filtering on each candidate item, retaining only those that truly satisfy the current query's needs.

In a study on a large corpus of randomized trials, a two-stage architecture (text-embedding recall + language model filtering) raised the recall of evidence needed for systematic reviews from 45% to 92%.

Language models can generate highly persuasive evaluative statements, but this does not constitute proof of the results themselves.

An Agent is essentially a transactional construct, that is, a function-calling loop wrapped around a language model that stops after executing a bounded task or waits for the next trigger.

To verify the transferability of the theory to neural network models, the researchers use a counterfactual proxy to estimate the influence of spurious-correlation dependence and use the generalized Gauss-Newton approximation to estimate the curvature of the language model.

Traditional language models usually split representation and prediction into two separate structures: representation is handled by the encoder or hidden layers, and prediction is performed by the output head.